Change the way we do mrt dumps in the rde (from the retarded only one dump

possible to as many as you like). The mrt dumps are now done with the same
tree walking code as the control messages. Makes the RDE respond a bit faster
and use less memory while dumping.

3 files changed, 110 insertions, 60 deletions

diff --git a/usr.sbin/bgpd/mrt.c b/usr.sbin/bgpd/mrt.c
index 31b2de8ec4e..bee49a670b7 100644
--- a/usr.sbin/bgpd/mrt.c
+++ b/usr.sbin/bgpd/mrt.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mrt.c,v 1.61 2009/06/04 22:08:19 claudio Exp $ */
+/*	$OpenBSD: mrt.c,v 1.62 2009/06/05 17:36:49 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org>
@@ -365,14 +365,6 @@ fail:
 	return (-1);
 }
 
-static u_int16_t sequencenum = 0;
-
-void
-mrt_clear_seq(void)
-{
-	sequencenum = 0;
-}
-
 void
 mrt_dump_upcall(struct rib_entry *re, void *ptr)
 {
@@ -386,14 +378,22 @@ mrt_dump_upcall(struct rib_entry *re, void *ptr)
 	 */
 	LIST_FOREACH(p, &re->prefix_h, rib_l) {
 		if (mrtbuf->type == MRT_TABLE_DUMP)
-			mrt_dump_entry(mrtbuf, p, sequencenum++,
+			mrt_dump_entry(mrtbuf, p, mrtbuf->seqnum++,
 			    p->aspath->peer);
 		else
-			mrt_dump_entry_mp(mrtbuf, p, sequencenum++,
+			mrt_dump_entry_mp(mrtbuf, p, mrtbuf->seqnum++,
 			    p->aspath->peer);
 	}
 }
 
+void
+mrt_dump_done(void *ptr)
+{
+	struct mrt		*mrtbuf = ptr;
+
+	mrtbuf->type = MRT_STATE_REMOVE;
+}
+
 int
 mrt_dump_hdr_se(struct buf ** bp, struct peer *peer, u_int16_t type,
     u_int16_t subtype, u_int32_t len, int swap)
diff --git a/usr.sbin/bgpd/mrt.h b/usr.sbin/bgpd/mrt.h
index c19a983f516..136d84cd814 100644
--- a/usr.sbin/bgpd/mrt.h
+++ b/usr.sbin/bgpd/mrt.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mrt.h,v 1.21 2009/06/04 22:08:19 claudio Exp $ */
+/*	$OpenBSD: mrt.h,v 1.22 2009/06/05 17:36:49 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org>
@@ -277,6 +277,7 @@ struct mrt {
 	u_int32_t		peer_id;
 	u_int32_t		group_id;
 	enum mrt_type		type;
+	u_int16_t		seqnum;
 };
 
 struct mrt_config {
@@ -302,6 +303,7 @@ void		 mrt_dump_state(struct mrt *, u_int16_t, u_int16_t,
 		     struct peer *);
 void		 mrt_clear_seq(void);
 void		 mrt_dump_upcall(struct rib_entry *, void *);
+void		 mrt_dump_done(void *);
 void		 mrt_write(struct mrt *);
 void		 mrt_clean(struct mrt *);
 void		 mrt_init(struct imsgbuf *, struct imsgbuf *);
diff --git a/usr.sbin/bgpd/rde.c b/usr.sbin/bgpd/rde.c
index e5acf6f9727..0a95b5677a4 100644
--- a/usr.sbin/bgpd/rde.c
+++ b/usr.sbin/bgpd/rde.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde.c,v 1.252 2009/06/04 22:08:19 claudio Exp $ */
+/*	$OpenBSD: rde.c,v 1.253 2009/06/05 17:36:49 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -38,7 +38,7 @@
 #define PFD_PIPE_MAIN		0
 #define PFD_PIPE_SESSION	1
 #define PFD_PIPE_SESSION_CTL	2
-#define PFD_MRT_FILE		3
+#define PFD_PIPE_COUNT		3
 
 void		 rde_sighdlr(int);
 void		 rde_dispatch_imsg_session(struct imsgbuf *);
@@ -76,6 +76,7 @@ void		 rde_dump_prefix_upcall(struct rib_entry *, void *);
 void		 rde_dump_prefix(struct ctl_show_rib_request *);
 void		 rde_dump_ctx_new(struct ctl_show_rib_request *, pid_t,
 		     enum imsg_type);
+void		 rde_dump_mrt_new(struct mrt *, pid_t, int);
 void		 rde_dump_done(void *);
 
 void		 rde_up_dump_upcall(struct rib_entry *, void *);
@@ -111,7 +112,6 @@ struct filter_head	*rules_l, *newrules;
 struct imsgbuf		*ibuf_se;
 struct imsgbuf		*ibuf_se_ctl;
 struct imsgbuf		*ibuf_main;
-struct mrt		*mrt;
 struct rde_memstats	 rdemem;
 
 struct rde_dump_ctx {
@@ -120,6 +120,14 @@ struct rde_dump_ctx {
 	sa_family_t			af;
 };
 
+struct rde_mrt_ctx {
+	struct rib_context	 ribctx;
+	struct mrt		*mrt;
+};
+
+struct mrt_head rde_mrts = LIST_HEAD_INITIALIZER(rde_mrts);
+u_int rde_mrt_cnt;
+
 void
 rde_sighdlr(int sig)
 {
@@ -146,12 +154,15 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 	struct passwd		*pw;
 	struct peer		*p;
 	struct listen_addr	*la;
-	struct pollfd		 pfd[4];
+	struct pollfd		*pfd = NULL;
 	struct filter_rule	*f;
 	struct filter_set	*set;
 	struct nexthop		*nh;
 	struct rde_rib		*rr;
-	int			 i, timeout;
+	struct mrt		*mrt, *xmrt;
+	void			*newp;
+	u_int			 pfd_elms = 0, i, j;
+	int			 timeout;
 
 	switch (pid = fork()) {
 	case -1:
@@ -210,7 +221,6 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 		LIST_REMOVE(mrt, entry);
 		free(mrt);
 	}
-	mrt = NULL;
 
 	while ((la = TAILQ_FIRST(config->listen_addrs)) != NULL) {
 		TAILQ_REMOVE(config->listen_addrs, la, entry);
@@ -245,8 +255,20 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 	}
 
 	while (rde_quit == 0) {
+		if (pfd_elms < PFD_PIPE_COUNT + rde_mrt_cnt) {
+			if ((newp = realloc(pfd, sizeof(struct pollfd) *
+			    (PFD_PIPE_COUNT + rde_mrt_cnt))) == NULL) {
+				/* panic for now  */
+				log_warn("could not resize pfd from %u -> %u"
+				    " entries", pfd_elms, PFD_PIPE_COUNT +
+				    rde_mrt_cnt);
+				fatalx("exiting");
+			}
+			pfd = newp;
+			pfd_elms = PFD_PIPE_COUNT + rde_mrt_cnt;
+		}
 		timeout = INFTIM;
-		bzero(pfd, sizeof(pfd));
+		bzero(pfd, sizeof(struct pollfd) * pfd_elms);
 		pfd[PFD_PIPE_MAIN].fd = ibuf_main->fd;
 		pfd[PFD_PIPE_MAIN].events = POLLIN;
 		if (ibuf_main->w.queued > 0)
@@ -264,11 +286,13 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 		else if (rib_dump_pending())
 			timeout = 0;
 
-		i = 3;
-		if (mrt && mrt->wbuf.queued) {
-			pfd[PFD_MRT_FILE].fd = mrt->wbuf.fd;
-			pfd[PFD_MRT_FILE].events = POLLOUT;
-			i++;
+		i = PFD_PIPE_COUNT;
+		LIST_FOREACH(mrt, &rde_mrts, entry) {
+			if (mrt->wbuf.queued) {
+				pfd[i].fd = mrt->wbuf.fd;
+				pfd[i].events = POLLOUT;
+				i++;
+			}
 		}
 
 		if (poll(pfd, i, timeout) == -1) {
@@ -301,13 +325,20 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 		if (pfd[PFD_PIPE_SESSION_CTL].revents & POLLIN)
 			rde_dispatch_imsg_session(ibuf_se_ctl);
 
-		if (pfd[PFD_MRT_FILE].revents & POLLOUT) {
-			mrt_write(mrt);
-			if (mrt->wbuf.queued == 0) {
+		for (j = PFD_PIPE_COUNT, mrt = LIST_FIRST(&rde_mrts);
+		    j < i && mrt != 0; j++) {
+			xmrt = LIST_NEXT(mrt, entry);
+			if (pfd[j].fd == mrt->wbuf.fd &&
+			    pfd[j].revents & POLLOUT)
+				mrt_write(mrt);
+			if (mrt->wbuf.queued == 0 && 
+			    mrt->type == MRT_STATE_REMOVE) {
 				close(mrt->wbuf.fd);
+				LIST_REMOVE(mrt, entry);
 				free(mrt);
-				mrt = NULL;
+				rde_mrt_cnt--;
 			}
+			mrt = xmrt;
 		}
 
 		rde_update_queue_runner();
@@ -320,6 +351,13 @@ rde_main(struct bgpd_config *config, struct peer *peer_l,
 	if (debug)
 		rde_shutdown();
 
+	while ((mrt = LIST_FIRST(&rde_mrts)) != NULL) {
+		msgbuf_clear(&mrt->wbuf);
+		close(mrt->wbuf.fd);
+		LIST_REMOVE(mrt, entry);
+		free(mrt);
+	}
+
 	msgbuf_clear(&ibuf_se->w);
 	free(ibuf_se);
 	msgbuf_clear(&ibuf_se_ctl->w);
@@ -530,9 +568,9 @@ rde_dispatch_imsg_parent(struct imsgbuf *ibuf)
 	struct rde_peer		*peer;
 	struct filter_rule	*r;
 	struct filter_set	*s;
-	struct mrt		*xmrt;
+	struct mrt		 xmrt;
 	struct nexthop		*nh;
-	int			 n, reconf_in = 0, reconf_out = 0;
+	int			 n, fd, reconf_in = 0, reconf_out = 0;
 
 	if ((n = imsg_read(ibuf)) == -1)
 		fatal("rde_dispatch_imsg_parent: imsg_read error");
@@ -672,37 +710,15 @@ rde_dispatch_imsg_parent(struct imsgbuf *ibuf)
 				log_warnx("wrong imsg len");
 				break;
 			}
-
-			xmrt = calloc(1, sizeof(struct mrt));
-			if (xmrt == NULL)
-				fatal("rde_dispatch_imsg_parent");
-			memcpy(xmrt, imsg.data, sizeof(struct mrt));
-			TAILQ_INIT(&xmrt->wbuf.bufs);
-
-			if ((xmrt->wbuf.fd = imsg_get_fd(ibuf)) == -1)
+			memcpy(&xmrt, imsg.data, sizeof(xmrt));
+			if ((fd = imsg_get_fd(ibuf)) == -1)
 				log_warnx("expected to receive fd for mrt dump "
 				    "but didn't receive any");
-			else if (xmrt->type == MRT_TABLE_DUMP ||
-			    xmrt->type == MRT_TABLE_DUMP_MP) {
-				u_int16_t id;
-
-				/* do not dump if another is still running */
-				id = rib_find(mrt->rib);
-				if (id == RIB_FAILED)
-					log_warnx("non existing RIB %s for mrt "
-					    "dump", mrt->rib);
-				else if (mrt == NULL || mrt->wbuf.queued == 0) {
-					free(mrt);
-					mrt = xmrt;
-					mrt_clear_seq();
-					rib_dump(&ribs[id], mrt_dump_upcall,
-					    mrt, AF_UNSPEC);
-					break;
-				} else
-					log_warnx("dump failed: already in progress");
-			}
-			close(xmrt->wbuf.fd);
-			free(xmrt);
+			else if (xmrt.type == MRT_TABLE_DUMP ||
+			    xmrt.type == MRT_TABLE_DUMP_MP) {
+				rde_dump_mrt_new(&xmrt, imsg.hdr.pid, fd);
+			} else
+				close(fd);
 			break;
 		case IMSG_MRT_CLOSE:
 			/* ignore end message because a dump is atomic */
@@ -1917,6 +1933,39 @@ rde_dump_done(void *arg)
 	free(ctx);
 }
 
+void
+rde_dump_mrt_new(struct mrt *mrt, pid_t pid, int fd)
+{
+	struct rde_mrt_ctx	*ctx;
+	u_int16_t		 id;
+
+	if ((ctx = calloc(1, sizeof(*ctx))) == NULL ||
+	    (ctx->mrt = calloc(1, sizeof(struct mrt))) == NULL) {
+		log_warn("rde_dump_mrt_new");
+		return;
+	}
+	memcpy(ctx->mrt, mrt, sizeof(struct mrt));
+	TAILQ_INIT(&ctx->mrt->wbuf.bufs);
+	ctx->mrt->wbuf.fd = fd;
+	ctx->mrt->type = MRT_STATE_RUNNING;
+	id = rib_find(ctx->mrt->rib);
+	if (id == RIB_FAILED) {
+		log_warnx("non existing RIB %s for mrt dump", ctx->mrt->rib);
+		free(ctx->mrt);
+		free(ctx);
+		return;
+	}
+	ctx->ribctx.ctx_count = RDE_RUNNER_ROUNDS;
+	ctx->ribctx.ctx_rib = &ribs[id];
+	ctx->ribctx.ctx_upcall = mrt_dump_upcall;
+	ctx->ribctx.ctx_done = mrt_dump_done;
+	ctx->ribctx.ctx_arg = ctx->mrt;
+	ctx->ribctx.ctx_af = AF_UNSPEC;
+	LIST_INSERT_HEAD(&rde_mrts, ctx->mrt, entry);
+	rde_mrt_cnt++;
+	rib_dump_r(&ctx->ribctx);
+}
+
 /*
  * kroute specific functions
  */
@@ -2758,7 +2807,6 @@ rde_shutdown(void)
 	attr_shutdown();
 	pt_shutdown();
 	peer_shutdown();
-	free(mrt);
 }
 
 int